The American Journal of Bioethics
Volume 15, Issue 4, 2015
Selecting the Right Tool For the Job
Arthur L. Caplana*, Carolyn Plunkettb & Bruce Levinc
Published online: 09 Apr 2015
There are competing ethical concerns when it comes to designing any clinical research study. Clinical trials of possible treatments for Ebola virus are no exception. If anything, the competing ethical concerns are exacerbated in trying to find answers to a deadly, rapidly spreading, infectious disease. The primary goal of current research is to identify experimental therapies that can cure Ebola or cure it with reasonable probability in infected individuals. Pursuit of that goal must be methodologically sound, practical and consistent with prevailing norms governing human subjects research. Some maintain that only randomized controlled trials (RCTs) with a placebo or standard-of-care arm can meet these conditions. We maintain that there are alternative trial designs that can do so as well and that sometimes these are preferable to RCTs.
If the goal of conducting trials in epidemic ravaged West Africa is to rapidly find an intervention that cures the infected and blunts the epidemic, then canonical RCT designs are not the only or even the best choice. The World Health Organization, Doctors without Borders, and other partners who coordinate trials on experimental agents agree (see Boseley 2014). There are practical reasons why placebo or SOC-controlled trials will be difficult if not impossible to undertake. It is particularly important to recognize that testing against the null hypothesis is neither appropriate nor necessary at this point in an out-of-control lethal epidemic. Instituting alternative clinical trial designs can provide useful information for the elimination or selection of prospective therapies. And that is what morally we owe those who are dying or at grave risk in environments where they have no other realistic means of survival.
This issue includes a number of Open Peer Commentaries on this Target Article and the author’s response below:
The Perfect Must Not Overwhelm the Good: Response to Open Peer Commentaries on “Selecting the Right Tool For the Job”
Arthur L. Caplana*, Carolyn Plunkettb & Bruce Levinc
Published online: 09 Apr 2015
We thank each of the peer commentators for their contributions. The key points they raise fall into five broad but interrelated categories of questions: (1) What are the appropriate goals of research on treatments for Ebola virus disease (EVD)? (2) Must we test the null hypothesis to achieve those goals? (3) What treatments and/or therapies should be included in trials? (4) What technical aspects of statistical design, implementation, and data analysis are most relevant to trials? (5) How should scarce research resources best be allocated to achieve the goal of reducing mortality from EVD?
Answers to questions 1 and 2 highlight strong disagreements between our critics and ourselves. We stated at the outset of our article (Caplan, Plunkett, and Levin 2015) that
The guiding methodologic question of clinical trials … in an epidemic that has spread out of control is not to test a “null hypothesis” that nothing works in carefully controlled circumstances but, rather, to assess among potentially promising agents, some of which have proven safety records, which stands the best chance of working. (5)
By “working” we mean, as Degeling and colleages say, working within “the socio-cultural, economic and political conditions in which it is likely to be used” (Degeling, Johnson, and Mayes 2015, 43). Many of our critics do not agree that this is an appropriate goal much less the appropriate goal. Dawson (2015) claims that the guiding question of research should be, “Can a new agent improve on our best treatment practices or not?” Similarly, Rid (2015) says, “If [Caplan and colleagues’ proposed randomized selection trial does] not answer the relevant question, namely, whether the experimental treatments add something to supportive care, their other advantages matter little.” Nelson and colleagues (2015) in various comments are in agreement with Dawson and Rid. All agree that a traditional randomized controlled trial (RCT) with a standard-of-care (SOC) control arm is the most expedient, and most ethical, way of proceeding.
We termed this strident defense of RCTs the “hypothesis test reflex.” To some of our critics, especially those writing from the perspective of the Food and Drug Administration (FDA), which has long been exceedingly wary of any other goal and method for testing novel drugs (Kurihara 2014) it appears to be beyond question that the primary (if not the only) goal of “research” with EVD patients must be to test whether an experimental treatment is safe and more effective than what is currently available to a high degree of certainty. Given that perspective, an alternative trial design that does not accomplish that goal cannot be deemed valid research, because it does not answer the stipulated “relevant question,” and, insofar as it has “little chance of providing interpretable efficacy and safety data,” yields an approach the critics believe to be useless and thus unethical.
There is a tautology in this line of reasoning: By definition, to qualify as research any research must test the null hypothesis; therefore, if a design such as a randomized selection trial (RST) does not do that, it must be faulty because it does not test the null hypothesis. We do not dispute the fact that an RST does not test the null hypothesis; we simply do not agree that our critics’ assumptions about the goals of research are of paramount importance in an uncontrolled, highly lethal epidemic. Thus, their proposed trials do not comprise the only valid design to meet what we think are more important goals—trying to save lives quickly while learning.
Given our goal—to assess among potentially promising interventions which one stands the best chance of working in West Africa in the midst of a deadly, fast moving epidemic—it is legitimate to start from the premise that the null hypothesis is false and get down to the business of finding the best agent to use. It also makes sense to follow our course since it is not clear local populations will accept RCTs—a point our critics fail to adequately engage (Adebamowo et al. 2014; Kupferschmidt and Cohen 2015).
In contrast to our critics, we think RSTs provide a better way to answer question 3: What treatments and/or therapies should be included in trials? Our answer is this: any treatment, therapy or intervention, or combination thereof, that stands a chance at being effective in the field, with the qualification that it has passed basic safety testing. We are sensitive to the fact that only 10% of drugs that enter Phase I safety testing are ultimately approved (Hay et al. 2014). Several commentators cited this fact in response to our claim that it is “reasonably plausible that at least some candidates are better than available SOC.” Though several commentators insist we must, we do not assume that interventions tested in the RST have “high ex ante probability” of being effective (Millum 2015) or will perform like “Babe Ruth” (Rid 2015). The interventions need only have a plausible biologic explanation for their proposed efficacy, appear to be safe in humans, be deliverable in the field, and be acceptable to potential recipients. Our assumption for initiating a trial is quite modest: At least one biologically plausible, safety-tested intervention may provide some benefit over the actual standard of care in most Ebola treatment units (ETUs). The morally essential goal is to rapidly identify the most promising agents under field conditions in trials that the population will find acceptable (Caplan 2015).
Among the proposed interventions that would be tested in an RST is “best available supportive care” (BASC), defined in Annette Rid’s piece as fluid replacement, broad-spectrum antibiotics, malaria treatment, and antipyretics. Contrary to our critics, we assume neither that BASC is already an effective treatment nor that it is widely available. It is one option among others, which may—or may not—be more effective than rival unproven or experimental treatments.
None of the writers appear to question that BASC is already an effective treatment compared to what the actual available level of supportive care was during the horrible outbreak in much of West Africa during the latter months of 2014 and continuing even today in parts of the region. We don’t hear the regulators who criticize us while continuing to plump for RCTs (Nelson et al. 2015) calling for a traditional RCT to test whether BASC is actually better than what was or is now available in most treatment locations in West Africa. Why then must we be so completely agnostic as to whether some other promising treatments such as convalescent blood products, monoclonal antibodies, or antiviral medicines might not work better than the care actually deliverable in most locations in the most impacted nations in West Africa? There is already preliminary evidence that favipiravir may reduce mortality by half in patients with low to moderate levels of Ebola virus, perhaps from 30% to 15% (Fink 2015), though we do not know what level of care study participants received in addition to the drug because study data have not yet been made available. Researchers are hopeful the easy-to-administer and readily available drug will also provide a benefit to children with EVD; the trial is set to expand to include children older than 1 year (Bouazza et al. 2015).
If one takes for granted, as our critics appear to do, that BASC is effective (compared to existing care in West Africa) and one further assumes that such care is widely available due to strengthened worldwide support, education of local populations, training of caregivers, and augmented distribution networks, then it should be given to all patients. In that case, we believe still that an RST should then be undertaken to determine the most promising additions to such care. One feature of selection procedures, which none of our commentators mention, is that if there were truly no differences in efficacy among all the candidate adjuvants to best available supportive care, then it would hardly matter which one or ones were selected. We would be free to select among the options based on practical concerns like deliverability, availability of health care workers, infrastructure, cost, local values, and so on. If all of them turned out actually to be worse than best available supportive care alone, this fact would become self-evident in short order as the nationwide case-fatality rates would fail to drop.
Unfortunately, we do not believe that BASC as defined by Rid is generally feasible, yet, throughout West Africa. Louis (2015) reminds us that the elements of BASC are still up for debate. Even intravenous rehydration and personal protective equipment may be unavailable in certain regions for practical reasons, as Waldman and Nieburg (2015) point out. Also, more advanced elements of supportive care such as kidney dialysis, biochemical monitoring, or ventilators, which are widely available in the United States and elsewhere in the developed world, are currently not a reality in West Africa. As such, BASC ought to be treated in West Africa like other experimental treatments whose efficacy appears promising but the value of whose widespread and rapid distribution is uncertain. Seen that way, evidence speaking to the regulator’s question of interest would become available even in an RST.
But to return to our main points, (i) even if BASC does reduce mortality from 70% to 30%, a treatment with a 30% chance of death is still a dire prospect to ask a patient to consent to, and (ii) we don’t think the primary goal of interventions should be to prove that something is better than BASC. The price to be paid for the holy grail of p < 0.05 is too high to those sick and dying (Caplan et al. 2015). We need to promptly identify the available treatments that will best save lives with useful evidence—the perfect ought not be the enemy of the good in the middle of a deadly epidemic.
Rid (2015) goes to some length to make our illustrative schoolyard baseball analogy work for null hypothesis testers. To extend the metaphor a bit more than she does, a professional baseball manager can surely conduct research on whether potential draftees may or may not improve the team’s strength, but let us not forget that the first priority is to quickly select those whom the manager thinks, based on incomplete evidence, are the best available players during the draft period.
Louis (2015), in addressing the fourth question, urges us to “expand the toolkit” rather than dwelling on differences of modes of statistical inference. We couldn’t agree more, though it would still be prudent to keep the trial designs simpler rather than more complex. Differences in statistical approaches and randomization schemes pale in importance compared to the goals of quickly trying interventions. In response to various comments on the technical aspects of statistical design, implementation, and data analysis, we recognize that we did not include an extended explanation of particular design details, in part because selection procedures have been known for over half a century and adaptive sequential selection procedures have also been studied for more than three decades. Our hope is that by discussing the issues, suggesting alternative randomized approaches in broad brush, and responding to our critics’ key points, we may help to broaden a discussion of the technical requirements for the kind of testing that ought to be done during epidemic crises and other humanitarian emergencies.
Several commentators, notably Millum, Degeling and colleagues, and Waldman and Nieburg, raised the fifth question: How should scarce research resources best be allocated to achieve the goal of reducing mortality from EVD? Millum says, “[Caplan and colleagues’] approach might make sense if there were no other ways in which the limited resources expended on distributing experimental therapies could be used to combat Ebola.” Waldman and Nieburg claim, “Socioanthropological research into issues like these may actually save more lives in the long run than attempts to find optimal treatments.” Degeling and colleagues say, “The development of an effective therapy for EVD is only part of the ‘the job’ that needs to be completed.” Ideally, there would be sufficient research funding to tackle Ebola from many angles.
We agree. Research into better personal protective equipment, cultural practices, and environmental issues, among other contributors to the current Ebola epidemic, would help reach the desirable goal of saving more lives and controlling this outbreak. But examining drug development makes sense too. It is a strategy worth pursing if we move away from the “hypothesis test reflex.” In deciding what is the best, most equitable way to attack epidemic diseases, trial design issues are crucial in deciding how best to allocate available resources.
We thank Dawson for reminding us of lessons learned from early AIDS trials. The proposed RST, if implemented, would be the first of possibly many steps toward finding a “slam dunk” treatment for Ebola, much as early AIDS trials were the first of many steps toward modern, highly effective, combination antiretroviral treatments. As we argue, the RST balances scientific rigor with ethical and practical concerns of providers, governmental and nongovernmental agencies, and local community members.
The issues about what type of trials to conduct are not going to go away. As the current Ebola outbreak subsides patients will (thankfully) become increasingly more difficult to find, let alone recruit to a traditional RCT. Until a successful preventative solution can be found, we must act as quickly as we can to find what best treats patients during the next outbreak and the next humanitarian emergency.